Menu
Generate and iterate images in Cursor using the built-in image model and strong prompts. Use when creating icons, illustrations, UI mockups, diagrams, marketing visuals, or any raster asset from a text description or reference image.
Systematic 4-phase debugging with root cause investigation. Use when fixing bugs to prevent random fixes.
Search and navigate large codebases efficiently. Use when finding specific code patterns, tracing function calls, understanding code structure, or locating bugs. Handles semantic search, grep patterns, AST analysis.
Apply appropriate design patterns (Singleton, Factory, Observer, Strategy, etc.) to solve architectural problems. Use when refactoring code architecture, implementing extensible systems, or following SOLID principles.
Technical documentation expert for creating clear, comprehensive documentation. Use when user asks to write docs, create README, or document code.
Writing clear, discoverable software documentation following the Eight Rules and Diataxis framework. Use when creating README files, API docs, tutorials, how-to guides, or any project documentation. Automatically enforces docs/ location, linking requirements, and runnable examples.
Build modern React/Next.js frontends. Use when creating web applications, choosing frontend stack, structuring components, or implementing UI/UX designs. Covers React, Next.js, Tailwind CSS, and component patterns.
| name | cursor-image-generation |
| description | Generate and iterate images in Cursor using the built-in image model and strong prompts. Use when creating icons, illustrations, UI mockups, diagrams, marketing visuals, or any raster asset from a text description or reference image. |
| metadata | {"author":"oh-my-cursor","version":"1.0.0"} |
Generate images in the Cursor agent using the GenerateImage tool. Image generation is backed by Google Nano Banana Pro. Previews are saved under assets/ by default unless you specify otherwise.
This skill is about prompting and workflow, not about replacing Figma or vector code (use other skills for those).
The user may give a short or vague request (“a hero for the login page”, “cyberpunk icon”). Do not pass that string raw to GenerateImage when it lacks the layers this skill describes. Instead:
The skill is the contract: the agent’s job is to expand and sharpen the user’s intent before generation, then iterate with deltas.
The bullets below are a condensed synthesis of common guidance for Nano Banana Pro / Gemini image models — not verbatim quotes. For authoritative wording and edge cases, use the References below.
They align with the spirit of Google’s public guides (prompt tips, Google Cloud guide, DeepMind prompt guide):
"cyber, 4k, hdr, epic") unless you deliberately want a tag-like aesthetic."bold geometric sans", "narrow serif for headlines"). Ask for legibility and high contrast if the text is important."change the background to warm beige", "make the logo 20% larger", "remove the extra person on the left") instead of a full new prompt.assets/ with a descriptive filename (e.g. assets/hero-spring-campaign.png).Square brackets [like-this] mark placeholders to fill in. Double quotes "..." mark exact text the model should render in the image.
App icon (square, legible at small size)
Square app icon, centered symbol of [subject], flat vector style with subtle depth, limited palette [colors], 10% safe margin from edges, no tiny text, crisp edges, high contrast on [background tone].
UI mockup still (marketing)
Photorealistic product screenshot of a [mobile/web] app, [screen name] view, centered device, soft studio lighting, neutral background, clean sans UI. Render the following text exactly: headline "[headline text]", button label "[CTA text]". Modern SaaS aesthetic.
Illustration (not photo)
Editorial illustration of [subject], [mood], limited palette [colors], visible brush texture or clean vector shapes (pick one), generous whitespace, no photorealistic faces unless requested.
Diagram / concept
Isometric diagram of [system], simple shapes, light grid, high contrast lines, no clutter, presentation slide style. Label the zones exactly: "[zone A label]", "[zone B label]".
| Weak | Stronger |
|---|---|
"A nice logo for my app" | Minimal wordmark for a productivity app, lowercase sans-serif feel, single accent color #2563EB on white, generous letter-spacing, no icon, horizontal logo lockup. |
"Cyberpunk city" | Wide 16:9 cinematic shot of a rainy cyberpunk street at night, neon reflections on wet asphalt, single vanishing point, shallow depth of field, no readable text, teal and magenta accents. |
"Fix the image" | Keep the same subject and composition; change only the background to soft gradient from #0f172a to #1e293b; leave lighting on the subject unchanged. |